News for hackers. Stuff that matters.
Re:Extracting concepts from GPT-4 by andy12_ on Thursday June 06, 2024 @04:20PM (#40601917)
a.k.a, the same work as Anthropic, but with less interpretable and interesting features. I guess there won't be Golden Gate[0] GPT anytime soon.

I mean, you just have to compare the couple of interesting features of the OpenAI feature browser [1] and the features of the Anthropic feature browser [2].

[0] https://twitter.com/AnthropicAI/status/1793741051867615494

[1] https://openaipublic.blob.core.windows.net/sparse-autoencode...

[2] https://transformer-circuits.pub/2024/scaling-monosemanticit...

Re:Extracting concepts from GPT-4 by leogao on Thursday June 06, 2024 @11:21PM (#40604992)
Note that we focus on random positive activations, which are less susceptible to interpretability illusions than top activations (but also look less impressive as a result). We also provide access to random uncherrypicked features, whereas Anthropic does not. We made these choices deliberately to give as accurate an impression of autoencoder feature quality as possible.

Also note that GPT-4 is a more powerful model than Sonnet, which makes it harder to train autoencoders with the same quality features.

Re:Extracting concepts from GPT-4 by justanotherjoe on Friday June 07, 2024 @03:24AM (#40606123)
yeah this one is much less presentable than Anthropic's work. It sure looks bad on them to be compared so poorly like this.